International Journal of Solids and Structures 84 (2016) 139146

Contents lists available at ScienceDirect

International Journal of Solids and Structures

journal homepage: www.elsevier.com/locate/ijsolstr

A generalized theory of elastodynamic homogenization

for periodic media
H. Nassar, Q.-C. He, N. Auffray
Modlisation et Simulation Multi-Echelle, Universit Paris-Est, MSME UMR 8208 CNRS, 5 bd Descartes, 77454 Marne-la-Valle, France

a r t i c l e

i n f o

Article history:
Received 13 August 2015
Revised 18 December 2015
Available online 4 February 2016
Keywords:
Homogenization
Willis theory
Generalized continua
Optical branches
Band gaps

a b s t r a c t
For periodically inhomogeneous media, a generalized theory of elastodynamic homogenization is proposed so that even the long-wavelength and low-frequency asymptotic expansions of the resulting effective (or macroscopic) motion equation can, approximately but simultaneously, capture all the acoustic
and some of the optical branches of the microscopic dispersion curve. The key to constructing the generalized theory resides in incorporating new kinematical degrees of freedom in conjunction with rapidly
oscillating body forces as microscopic and macroscopic loadings while satisfying an energetical consistency constraint reminiscent of HillMandel lemma. By this constraint, an effective displacement eld is
naturally dened as the projection of a microscopic one onto the dual to the space of body forces. To
illustrate these results, a two-phase string is studied in detail.
2016 Elsevier Ltd. All rights reserved.

1. Introduction
The elastodynamic homogenization approaches reported up to
now in the literature are observed to run into diculties when being used to model dynamical effects over a wide frequency range.
1. The classical lowest-order Long-Wavelength (LW) LowFrequency (LF) homogenization approaches (Bensoussan et al.,
1978; Sanchez-Palencia, 1980) yield a homogeneous substitution Cauchy medium which misses all dispersive effects and all
internal resonances, i.e., all optical oscillation modes.
2. The higher-order LW-LF asymptotic homogenization approaches
(Andrianov et al., 2008; Boutin and Auriault, 1993) lead to effective strain-gradient media which can model well dispersive
behaviors and size effects but are valid only near the acoustic
branches independently of the order of the asymptotic approximations used.
3. The high-frequency asymptotic approaches (Antonakakis et al.,
2014; Boutin et al., 2014; Colquitt et al., 2014; Craster et al.,
2010; Daya et al., 2002; Nolde et al., 2011) are successful in
capturing high-frequency optical modes but still valid only in
the vicinity of some nite frequency.
4. The high-contrast asymptotic approaches (Auriault and Bonnet, 1985; Auriault and Boutin, 2012; Smyshlyaev, 2009) have a
wide frequency validity domain englobing an innite number of

Corresponding author. Tel.: +33 160967794.

E-mail address: hussein.nassar@univ-paris-est.fr (H. Nassar).

http://dx.doi.org/10.1016/j.ijsolstr.2016.01.022
0020-7683/ 2016 Elsevier Ltd. All rights reserved.

optical branches. However, the corresponding effective behavior

is complex and nonlocal in time.
5. The non-asymptotic theory of Willis (1997, 2011) yields exactly
the whole dispersion curve. Nonetheless, the described effective
elds are only relevant for low frequencies (Nassar et al., 2015b;
Srivastava and Nemat-Nasser, 2014).
The main purpose of the present paper is to construct a generalized theory of elastodynamic homogenization for periodic media
which improves the quality of the Willis effective behavior as an
approximation to the microscopic behavior in a way that LW-LF
asymptotic expansions become able to capture, approximately but
simultaneously, all the acoustic and some of the optical branches
of the microscopic dispersion curve. To achieve this purpose, new
kinematical Degrees Of Freedom (DOFs) are taken into account so
as to describe some short-wavelength components of the microscopic displacement eld which become dominant at high frequencies. The new DOFs are excited by incorporating various rapidly
oscillating body forces on the microscale and on the macroscale
under an energetical consistency constraint hereafter called Energy Equivalency Principle (EEP). The EEP is a balance between the
microscopic and macroscopic virtual works and is later proven to
yield a generalized version of the well-known HillMandel lemma.
With respect to Willis theory, we underline two major differences.
First, the incorporated loadings are much richer than those employed by Willis (1997, 2011). This has the consequence of reducing the error committed during the upscaling process and providing an extended frequency validity domain. Second, the EEP concerns virtual works and not their expectancies. From the physical

140

H. Nassar et al. / International Journal of Solids and Structures 84 (2016) 139146

standpoint, this leads to a clear distinction between the macroscale

and the microscale in terms of wavelengths. Nevertheless, it should
be pointed out that the generalized theory presented here is by
construction limited to periodically inhomogeneous media while
Willis theory is formally valid both for periodically and randomly
inhomogeneous media.
The paper is organized as follows. In Section 2, we recall some
geometrical elements useful for describing periodic media, summarize the equations governing the kinematics and dynamics of them,
and simplify these equations by using Bloch-wave expansions. The
main body of the generalized theory is presented in Section 3. The
EEP is rst postulated; the space of admissible body forces is then
dened as the set of macroscopically applied loadings; the effective displacement eld associated to a microscopic displacement is
obtained by the EEP and proven to be an improvement over the
one dened by Willis; the effective motion equation is nally derived in a formal way and a HilMandel relation is demonstrated.
In Section 4, an analytical LW-LF asymptotic approximation to the
effective motion equation is given for a particular 1D two-phase
string. Exact and approximate dispersion curves are plotted and
compared. It appears then how the resulting asymptotic model,
though based on LF expansions, can simultaneously capture acoustic and optical branches while conserving a low-order local motion
equation.
2. Preliminaries
In this section, some geometrical elements useful for the study
of periodic media are recalled. The governing equations of linear
elasticity are recapitulated. Bloch-wave expansions of elds and
work are also introduced.
2.1. Geometry and periodicity
Let  be a d-dimensional innite body. Dene E as the vector
space of translations acting on the points of . Given d independent translations (b j ) j=1... d , denote by R the subset of E obtained
by integer combinations of these vectors. The subset R is called
a lattice. Then, a scalar, vector or tensor eld h dened over  is
said to be R-periodic if and only if it satises h(x + r ) = h(x ) for
all points x  and all translations r R. Accordingly, h needs
being dened only over a unit cell



d


T = xo + rr =
r j b j , 1/2 r j < 1/2 ,
j=1

where xo , its center, is an arbitrary point of . Note that while

R-periodicity is well dened, the choice of bj and T is not unique.
Symbolize by E the dual space of E . A wavenumber k E acting on a translation r E produces a phase shift k r where () is
the usual dot product. Now, points of  and vectors of E can be
identied after choosing some origin xo . In what follows, we drop
xo so as to write k x instead of k (x xo ) for simplicity. The reciprocal lattice R of the direct lattice R is dened as the subset
of E consisting of wavenumbers such that ei x is R-periodic,
with i2 = 1. Also of interest is the rst Brillouin zone T dened
as the set of wavenumbers closer to the null wavenumber than to
any other wavenumber of R , i.e.,

T = kE

|k < k , R {0} .

This zone is uniquely dened and independent of T.

A function h dened over  can be expanded into plane waves
over E such that

h (x ) =

k eikx dd k.
h

In particular, when h is R-periodic, it can be written as the Fourier

series

h (x ) =

ei x .
h

Having this in mind, with respect to R, T can be seen as the support of slowly varying elds. In particular, among R-periodic functions, only constants have their wavenumber contained in T , i.e.,
T R = {0}.
Finally, call a Bloch wave, of wavenumber k and amplitude
k (x ), a function h (x) of the form
h
k

k (x )eikx ,
hk ( x ) = h
k (x ) is R-periodic.
where h
2.2. Constitutive and motion equations
Letting u(x, t) be the displacement vector for a point x  at
instant t, the strain eld and velocity eld v are derived according to

= s u, v = u ,
where is the space gradient operator, denotes the tensor
product, the superscripted s indicates symmetrization and a superscripted dot symbolizes differentiation with respect to time.
The stress tensor and momentum density p are then given by
the local constitutive equations of :

= C : ,

p = v,

with C and being the elastic stiffness tensor and the scalar mass
density, respectively, and the colon (:) standing for double contraction.
The motion equation of  reads

+ f = p
where ( ) is the divergence operator and f is a eld of externally
applied body forces. We shall mostly work with harmonic elds
of frequency . Therefore, all time derivatives can be substituted
by i-multiplications and time dependency can be dropped henceforth. The motion equation of  becomes the Helmholtz equation


C (x ) : s u(x ) + f (x ) = 2 (x )u(x )

(2.1)

where we have displayed x-dependencies and omitted dependencies.

In this work, the homogenization of  amounts to nding the
motion equation, hereafter called effective motion equation, of a
homogeneous medium substituting the initial inhomogeneous one,
under an energy equivalency constraint to be specied.
2.3. Bloch-wave expansions
The superposition principle makes it possible to work with elementary, such as plane-wave, body forces instead of arbitrary ones
f(x). It is however more convenient, for reasons that will become
clear, to work with Bloch-wave body forces. Then, let fk (x) be an
element of the Bloch-wave expansion of f(x) such that

f (x ) =

f k ( x ) dd k

fk (x )eikx dd k,

(2.2)

where fk (x ) is R-periodic and the symbol stands for equality by

denition.
For a given k T , the motion equation for a Bloch-wave body
force takes the form


C (x ) : s uk (x ) + fk (x )eikx = 2 (x )uk (x ).

H. Nassar et al. / International Journal of Solids and Structures 84 (2016) 139146

141

We now assume that C and are R-periodic so that the solution

uk (x) can be written as

uk (x ) = u k (x )eikx
with an R-periodic amplitude u k (x ) (Gazalet et al., 2013). In other
words, to the expansion (2.2), there corresponds a similar Blochwave expansion of the solution to (2.1):

u (x ) =

uk ( x ) dd k =

u k (x )eikx dd k,

(2.3)

with uk (x) being the displacement eld over  subjected to fk (x).

In contrast, except for a homogeneous body, there are no similar
results for plane waves.
Given the Bloch-wave form of uk , and simplifying the phase
factor eik x , the motion equation becomes, in terms of u k (x ) and
fk (x ),

( + ik ) {C (x ) : [( + ik ) s u k (x )]} + fk (x ) = 2 (x )u k (x ).
(2.4)
Last, the x-dependencies of Bloch amplitudes for displacements
and body forces were explicitly annotated. In what follows, all
elds are understood to be x-dependent unless otherwise stated.
2.4. External work
Dene the virtual work of external body forces f associated
with a virtual displacement eld u by

f u

where a superscripted denotes complex conjugation. Also, introduce the averaging operator  by

h =

1
|T |

h ( x ) dd x

for all R-periodic functions h. Note that  is independent of

the choice of T and is ill-dened for non-R-periodic functions.
Plancherels identity for Fourier transform and Parsevals identity
for Fourier series deliver then a similar identity for Bloch-wave expansions:

f u = ( 2 )d
=

( 2 )d
|T |

fk u k dd k

 
T

fk (x ) u k (x ) dd x dd k.

(2.5)

Consequently, in dening the energetically equivalent effective behavior, we can work with elds of a single Bloch wavenumber k
T and then apply the superposition principle even though work is
quadratic and not linear.

Fig. 1. The effective displacement eld D associated to a given microscopic one u is

geometrically interpreted as the projection of the latter onto the space of admissible
displacements. Spaces F and F are isomorphic and, here, are taken to be equal
up to a change in units.

3.1. Energy equivalency

In classical static or quasi-static homogenization, an energy
equivalency relation, known as HillMandel lemma, is proven for a
family of boundary conditions prescribed on a representative volume element as macroscopic loadings. Once the boundary conditions have been specied, HillMandel lemma can be used to
dene, by duality, the macroscopic stress in a strain-based approach or the macroscopic strain in a stress-based approach. In the
present formulation, admissible body forces applied globally to 
instead of boundary conditions are taken to be macroscopic loading. Then, an EEP is postulated so as to dualize body forces and
displacements. This duality will allow us to dene the macroscopic
displacement eld, called D, in terms of the microscopic one u,
once admissible body forces have been imposed.
Let F be the space of Bloch amplitudes fk , involved in (2.2),
of admissible body forces. The elements of F are seen as external
loadings likely to be applied to . Note that they will remain the
same after the scale transition. The space F acts as a parameter of
the approach to be elaborated and needs to be chosen adequately.
Next, let F , the space dual to F , be the space of Bloch amplitudes D k of admissible effective displacement elds. For a given
microscopic displacement eld u, the corresponding effective (or
macroscopic) displacement eld is dened as the unique admissible displacement such that

f D =

f u

(3.1)

for all admissible virtual body forces f (i.e., fk F , for all k T ).

Physically, the EEP (3.1) can be interpreted as requiring that the effective displacement eld D associated to a given microscopic eld
u be such that the work done by every admissible virtual body
force f in the course of D is equal to the one done by f in the
course of u. Geometrically, the EEP (3.1) simply means that D is the
projection of u onto the space of admissible displacements (Fig. 1).
On the basis of the EEP (3.1), a generalized HillMandel lemma will
be proven in Section 3.4.
Using the Bloch decomposition (2.5), the EEP (3.1) can be equivalently written in terms of Bloch amplitudes as

3. A general theory

k T , fk F ,

In this section, the energy equivalency principle (EEP), the corner stone of the present approach, is rst postulated, given a simple form and exploited to dene the effective displacement eld.
A formal derivation of the effective motion equation is then presented. The effective constitutive behavior is nonlocal in both space
and time which raises questions about its uniqueness (Fietz and
Shvets, 2010; Willis, 2011). In order to avoid this diculty, we
will be interested only in the effective motion equation which is
unique. Nonetheless, we will derive expressions for the generalized stress, momentum, velocity and strain measures which are, in
particular, needed for determining an effective constitutive law.

3.2. Effective displacement eld

fk D k = fk u k .

(3.2)

Bearing in mind the EEP (3.1), choosing the space F of admissible body forces becomes a key step toward elaborating a generalized theory. The choice of F depends ultimately on the degree of
accuracy with which D is required to approximate u. The bigger F
is, the closer D is to u. When all body forces are considered as admissible, the relation (3.1) implies D = u and the effective medium
is trivially the original one. In what follows, we study the rather
general case of practical importance where F is nite-dimensional
and show how D derives from (3.1) correspondingly.

142

H. Nassar et al. / International Journal of Solids and Structures 84 (2016) 139146

3.2.1. Admissible body forces

Given N linearly independent, k- and -independent, Rperiodic vector elds i with i = 1. . . N, a Bloch-wave body force
eld fk is admissible if and only if it has an R-periodic Bloch amplitude fk of the form

fk (x ) =

N


fki i (x ),

(3.3)

i=1

where the fki are constants. The space F is therefore of dimension

N.
With no loss of generality, let the subset (i )i=1...d , where d is
the dimension of , be formed of constant vectors and constitute
a basis for E . We call Fk the constant component of fk and write

fk (x ) =

d


fki i +

i=1

N


fki i (x ) Fk + fk (x ).

(3.4)

i=d+1

Above and from now on, the repeated Greek indices are understood to be summed over from d + 1 to N whereas the Latin ones
run from 1 to N unless otherwise specied. Integrating with respect to k over T , we obtain the generic form of admissible body
forces:

f ( x ) = F ( x ) + f ( x ) ( x ) .
Of most importance is the fact that elds F and f have their
supports contained in T . As such, they have wavelengths at least
twice as large as the characteristic length of a unit cell. Consequently, the DOFs F and f of admissible body forces are said to be
macroscopic. These DOFs are carried by R-periodic shape functions, the , describing the ways in which f can vary on the microscale. For example, taking N = d, we have f (x ) = F (x ) implying that body forces are not allowed to vary on the microscale. As
another example, setting N = d + 1 and d+1 (x ) = (x )e, where e
is a vertically oriented vector, we have f (x ) = F (x ) + f d+1 (x ) (x )e
so that the admissible variations of body forces on the microscale
are gravitational.
3.2.2. Effective displacement eld by the EEP
From now on, we assume that the form an orthonormal basis of F so that


i j = i j ,

i, j {1, . . . N},

where ij is the Kronecker delta. For {1, . . .d}, being constant entails

{d + 1, . . . N},   = 0,
meaning that being orthogonal to a constant is equivalent to having a zero average.
Injecting (3.4) in the expression of the virtual work, we obtain

fk u k = Fk u k + fk u k

(by constancy)
(by denition (3.5))

= fk Uk + uk

with

Uk u k ,

uk

(by orthogonality)

 


u k .

Then, it follows from (3.2) that

D k (x ) = Uk + uk (x ).

 

The above expression of the effective displacement eld results

from the EEP combined with a particular choice of admissible body
forces. It contains the classical translational displacement vector
U and additional generalized displacements u carried by the
shape functions . Once more, the shape functions dene the
way in which D varies on the microscale whereas the slowly varying DOFs U and u describe how D varies on the macroscale.
Willis (2011) proposed a homogenization theory in which shape
functions are taken to be i (x ) = w(x )ei , for i {1, . . .d}, where
w(x ) is a xed R-periodic function and the ei form a basis for E .
Taking w 1 yields the unweighted theory of 1997 (Willis, 1997)
and amounts to taking f = F and D = U . Here, we combine both
the weighted and unweighted Willis theories and use even more
general shape functions. As a consequence, D is a better approximation of u than U as will be seen in more detail.
3.2.3. Effective displacement eld through error minimization
First of all, rewriting (3.2) in the equivalent form

fk F ,

fk (u k D k ) = 0,

it is clear that u k D k is orthogonal to F and D k acts as the orthogonal projection of u k onto F (Fig. 1). Using the Pythagorean
theorem, it is easy to see that for any R-periodic eld h F ,


(u k h ) (u k h )  = (u k D k ) (u k D k )


+ (D k h ) (D k h )


(u k D k ) (u k D k ) .
Thus,

D k = argmin (u k h ) (u k h ) .

(3.6)

hF

This shows that the effective displacement Bloch amplitude is

the best admissible approximation to the microscopic one. Consequently, the effective displacement eld D, associated to a microscopic displacement eld u, can be seen as the best admissible
approximation to u. Note that this global optimal argument denition (where the support is ) is different from the local one introduced elswhere (Forest, 2006; Forest and Sab, 1998) (where the
support is a representative volume element) despite an apparent
resemblance.
The preceding denition of the effective displacement eld concretizes the intuition that the richer the DOFs of the generalized
substitution medium are, the closer D is to u:

if F1 F2
hF2

= Fk Uk + fk uk

D ( x ) = U ( x ) + u ( x ) ( x ) .

then min (u k h ) (u k h )  min (u k h ) (u k h ) .

(by orthogonality)

= Fk u k  + fk  u k 

= Fk Uk + fk uk

Finally, summing over T , it comes that

(by orthogonality)

(3.5)

hF1

In this sense, the generalized substitution medium to be obtained

by our theory is more realistic than the Willis substitution medium
in the above minimal error sense, at the cost of an additional kinematical complexity (see Fig. 2).
A remark is now in order. We have used a scalar product on
the space of body forces twice up till now: once to identify F
and F and once to orthonormalize the set of shape functions.
This scalar product is not unique and can be modied by adding
a weighting function such as mass density for instance. Note that
such choice has inuence neither on the denition of the DOFs u ,
nor on the effective motion equation to be found. It simply changes
the above quadratic error function and determines the mapping
(ui )i=1... N  D.

H. Nassar et al. / International Journal of Solids and Structures 84 (2016) 139146

143

Fig. 2. Plots of the real parts of the microscopic (u), Willis (U ), and generalized (D ) Bloch displacement amplitudes over one period for 3 eigenmodes: (k = 0, = 1 (0 ))
(top), (k = /2a, = 2 ( /2a )) (middle), (k = 0, = 2 (0 )) (bottom). Two shape functions have been used: a constant and a sine wave. Details are given in Section 4.

3.2.4. Effective displacement eld under innite scale separation

It is of interest to examine what D becomes under the hypothesis of innite scale separation, namely, when k 0 and 0.
It is known that in this case, to the lowest order, the displacement
eld depends only on the slow variable (refer to Boutin and Auriault, 1993, for instance). In terms of Bloch amplitudes, this means
that u k is constant. Consequently,

second order Green operator. Then,

Uk = u k  = u k ,

which, combined with (3.5), delivers the following expressions for

the components of the macroscopic displacement eld:

uk =

 u k  =   u k = 0.

Therefore, the translational DOF U is the only non-null component

of D, to the lowest order. The use of a generalized kinematics is
hence justied only under weak scale separation or for high frequencies when microscopic deformation modes become signicant.
Otherwise, it is enough to keep track of U exclusively as in the unweighted Willis theory. As a matter of fact, it has been observed
that a periodic medium was homogenizable in the Willis sense
over the acoustic and the rst optical branches only (Nassar et al.,
2015b; Srivastava and Nemat-Nasser, 2014). For higher frequencies,
one needs to use non-uniform shape functions.
3.3. Effective motion equation
Having specied body forces, the motion Eq. (2.4) becomes

( + ik ) {C (x ) : [( + ik ) s u k (x )]} + Fk + fk (x ) (x )
= 2 (x )u k (x ),
which needs to be solved over . Since u k is R-periodic, it is
enough to solve the above equation over a unit cell T under periodic boundary conditions. Let gk be the corresponding periodic

u k (y ) =
=

1
T

1 
T

gk (y, x ) fk (x ) dd x

gk (y, x ) dd x Fk +

1 
T

gk (y, x ) dd x fk
(3.7)

Uk = gk (y, x ) Fk + gk (y, x ) (x ) fk ,

uk =




(y ) gk (y, x ) Fk + (y ) gk (y, x ) (x ) fk ,

(3.8)

where  means averaging with respect to both x and y.

The Green operator of the effective medium Gk is given by the
last two equalities which can be written concisely as

ukj = Gkji fki ,

where no distinction is made between the classical and generalized

j
DOFs (recall that Uk = dj=1 uk j ). Inverting the preceding equation delivers the effective motion equation in Fourier domain:

Zki j ukj = fki ,

(3.9)

where Zk , the inverse of Gk , is called the effective impedance. It

depends implicitly on the frequency . By summing over k T
and over , we obtain the effective motion equation in x and t
as

Z i j ( x, t ) u j ( x, t ) = f i ( x, t ),
where Z(x, t) is an integro-differential operator and denotes convolution product with respect to space and time. The effective motion equation is hence nonlocal in both space and time and in-

144

H. Nassar et al. / International Journal of Solids and Structures 84 (2016) 139146

volves long wavelengths (k T ) only. This generalizes equation

(3.28) derived by Willis (1997) for periodic media.
3.4. Internal work

Fig. 3. Unit cell.

As mentioned earlier, it is not essential for achieving the main

purpose of the present work to derive an explicit expression for
the underlying effective constitutive law which is not unique as
in the theory of Willis. However, it is of interest to specify the
macroscopic stress, momentum, strain and velocity measures that
an effective constitutive law involves. In addition, these generalized
macroscopic measures will be shown to be related to their microscopic counterparts through an extended HillMandel relation.
3.4.1. Generalized stress and momentum measures
Said measures are taken to be the ones involved in the effective
motion equation written as a conservation law equivalent to (3.9).
Starting with the microscopic motion equation

( + ik ) k + Fk + fk = i p k ,

(3.10)

where k and p k are the Bloch amplitudes of stress and momentum, we take its volume average over a unit cell to obtain, with
the help of the divergence theorem,

k + Fk = iPk .
ik 

(3.11)

This is the rst effective motion equation involving the classical

macroscopic stress and momentum measures:

k  k ,


Pk  p k .

Substituting body forces by the corresponding stress and momentum measures according to (3.12) delivers


k : (iks Uk ) Pk (iUk )
k : k p k v k = 
+ k (ikuk ) pk (iuk ) sk uk .

Summing over k, and using Placherels identity, we obtain a generalized version of the HillMandel lemma:

{ : p v } =

{ : ( s U ) P (iU )

+ ( u ) p (iu ) s u }. (3.14)

This result is valid for all virtual couples ( , p) equilibrated by

an admissible body force eld and for all couples (, v) derived
from an arbitrary displacement eld u. From the above relation,
we identify the classical macroscopic measures of strain and velocity as s U and iU while the generalized ones are u , u
and iu .
Finally, the constructed macroscopic elds meet the most fundamental requirements for them to be interpreted as stresses, momenta, strains and velocities since they rigorously satisfy local balance and compatibility equations. Nonetheless, how to physically
interpret and measure these quantities in a precise way ultimately
depends on the chosen set of shape functions. More details are
given in Section 4.4 regarding this aspect.

Further, projecting Eq. (3.10) onto the space spanned by the other
shape functions gives rise to

4. An application: high-frequency behavior through

low-frequency asymptotics

ik

At this point, we have generalized Willis theory by using

enriched kinematics to improve the quality of approximation of
a microscopic displacement u by a macroscopic one D. The cost
is however the increasing complexity of the resulting effective
motion equation. A numerical procedure dedicated to the implementation of Willis theory or our previous one is quite heavy,
the effective behavior being nonlocal in both space and time with
innite radii of inuence in general.
Taylor asymptotic expansions provide an ecient way to approximate the nonlocal behavior with a local one under appropriate assumptions on k and . LW-LF expansions have the main
advantage of only requiring the solution of static problems but
present the disadvantage of being limited to the LF behavior. The
purpose of this section is to show explicitly how generalizing
Willis theory makes it possible to extend the validity domain of
LW-LF expansions to high-frequency behavior over a simple 1D example.





k s : k + fk = i p k ,

where, for simplicity, we have assumed the continuity of so

that the boundary term vanishes. The generalized stress and momentum measures can be identied as



k k , sk s : k ,

p k


p k .

The additional motion equation becomes then simply

sk + ik k + fk = i p k .

(3.12)

Note that Eqs. (3.11) and (3.12) on one hand, and (3.9) on the other,
are related to one another through a non-unique effective constitutive law whose characterization is beyond the purpose of the
present work (see the discussion by Willis, 2011, 2012).
In summary, the motion equations in the space domain are
given by

 + F = iP ,
s + + f = i p .

(3.13)

These equations are a micromechanical version of the equations

of equilibrium phenomenologically derived by Germain (1973).

4.1. Setting
Consider the periodically inhomogeneous string whose unit cell
is depicted in Fig. 3, and dene the shape function

(x ) = 2 sin ( x/a )
3.4.2. Generalized strain and velocity measures
Said measures are obtained by duality. The virtual work theorem combined with the EEP yields

which describes the rapidly oscillating body force



k : k p k v k = fk u k = Fk Uk + fk uk ,

and will carry the new DOF . The macroscopic displacement D

then reads

where k and v k are the Bloch amplitudes of the strain and velocity elds, respectively, given by

D = U +

k = ( + ik ) s u k , v k = iu k .

U = u,

f = F + q

with

=  u.

H. Nassar et al. / International Journal of Solids and Structures 84 (2016) 139146

145

4.2. LW-LF effective motion equation

The microscopic motion equation reads

( + ik ){C[( + ik )uk ]} + Fk + qk = 2 uk .
Instead of calculating the effective impedance Z for arbitrary k and
, we are interested here in a LW-LF Taylor expansion of Z, as k
0 and 0, which is straightforward to obtain by solving
a hierarchy of static motion equations. The hierarchy is obtained
by injecting an expansion of uk , in powers of k and , into the
above equation. This procedure is well described in the literature
(Andrianov et al., 2008; Boutin and Auriault, 1993; Smyshlyaev and
Cherednichenko, 20 0 0) and is skipped here. Calling ci and i the
stiffness and mass density of phase i for i {1, 2}, and a the halflength of a unit cell, the approximate effective impedance Z, truncated at order 2 in k and , is given by
c 1 c 2 2 1 + 2 2
ZUU = 2
k
,
c1 + c2
2

4 2 c1 c2 ( c1 c2 ) 2
2
Z U = ZU =
k
+
( 2 ) 2 ,
( c1 + c2 )2
1
Z =

order approximation by the present theory (two branches (1(2,2) ), in blue) and to
its fourth-order approximation by the present theory (two branches (1(4,2) ), in red).
(For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article.)

2 2 c1 c2
a2 c1 + c2
2 [( 2 6 )2 +21 ]c12 + 4(1 +2 )c1 c2 +[( 2 6 )1 +22 ]c22 2
2

( c 1 +c 2 ) 2

2 c1 c2 [(3 2 8 )c12 + 2(3 2 + 8 )c1 c2 + (3 2 8 )c22 ] 2

k ,
2
( c1 + c2 )3

with the approximate effective motion equation being

U
ZkUU Uk + Zk k = Fk ,

Zk Uk + Zk k = qk .

In the real domain, the above equation takes the form

c1 c2
1 + 2 4 2 c1 c2 (c1 c2 ) 
2
U
2
U  +

( 2 ) = F ,
c1 + c2
2
( c1 + c2 )2
1

2
2 2 c1 c2
4 2 c1 c2 (c1 c2 ) 

U
( 2 )U + 2

( c1 + c2 )2
1
a c1 + c2
+
+

2 [( 2 6 )2 + 21 ]c12 + 4(1 + 2 )c1 c2 + [( 2 6 )1 + 22 ]c22

2
( c1 + c2 )2
2 c1 c2 [(3 2 8 )c12 + 2(3 2 + 8 )c1 c2 + (3 2 8 )c22 ] 
= q,
( c1 + c2 )3

where a superscripted dot denotes

/ x.

/ t and the prime symbol means

4.3. Exact and approximate dispersion curves

 


( c 1 1 + c 2 2 ) 2
cos(2ka ) =
cos ( 1 /c1 + 2 /c2 )a

4 c 1 1 c 2 2

 


( c 1 1 c 2 2 ) 2

cos ( 1 /c1 2 /c2 )a .

4 c 1 1 c 2 2

(4.1)
The approximate dispersion curve is derived from the approximate
effective impedance according to

ZUU Z Z U ZU = 0.
The rst two branches of the exact and approximate dispersion
curves are drawn in Fig. 4 for the following arbitrary numerical
values of the string parameters

c2 = 100,

sistatic approximation (1(0 ) ), enriching the kinematics allows for

capturing the rst optical branch and grants a larger validity domain for LF asymptotic expansions. The approximate dispersion
branches of order 4 are more precise and almost indistinguishable
from the exact ones. The corresponding approximate dispersion relation was numerically found and is not given here. However, we
presented the resulting dispersion branches so as to get a glimpse
of the convergence rate of the asymptotic scheme. Note that simultaneously capturing additional optical branches requires including
richer body forces.
Why is it possible that LF Taylor expansions lead to a correct estimate of some optical modes? Physically speaking, for
high frequencies, inertial forces become important and shift the
energy carried by displacements toward shorter wavelengths.
Correspondingly, including rapidly oscillating body forces have two
benets. First, they simulate the effects of inertial forces. Second,
and most importantly, they oblige the macroscopic displacement
eld to include some short-wavelength components, necessary
for approximating the high-frequency behavior. Mathematically
speaking, including additional DOFs delays the appearance of some
singularities and extends the convergence domain of the LF Taylor
expansions (see Nassar et al., 2015a).

4.4. On the choice of shape functions

The expression of the exact dispersion curve is known and was

derived elsewhere (Andrianov et al., 2008). It reads:

c1 = 1 ,

Fig. 4. Exact dispersion curve (two branches (1, 2 ), in black) compared to its classical quasistatic approximation of order 0 (one branch (1(0) ), in green), to its second-

1 = 1 , 2 = 5 , a = 1 .

The plots are properly normalized so that units become irrelevant for our purposes. On Fig. 4, with respect to the classical qua-

The above method and results are by no means universal. Depending on the underlying microstructure and on the targeted frequency range, adequate shape functions can be chosen. Regarding
how to appropriately choose shape functions, the following comments are in order:
1. In our 1D example, the sinusoidal shape function was chosen
based on considerations similar to the ones arising in Braggs
reection where the rst term added to the coherent wave is
another Fourier component of the exact scattered (microscopic
in our terminology) eld (see, e.g., Qur, 1988).
2. In some situations, by inspecting the phases connectedness
and contrasts, an asymptotic analysis allows constructing shape
functions as particular quasistatic rst-order solutions (see, e.g.,
Auriault and Bonnet, 1985).
3. It can also be proven that including the nth periodic optical
oscillation mode as a shape function guarantees that the nth
optical dispersion branch is correctly approximated at k = 0.
In fact, this amounts to combining the classical quasistatic

146

H. Nassar et al. / International Journal of Solids and Structures 84 (2016) 139146

homogenization theory with the high-frequency homogenization theory suggested by Daya et al. (2002) and Craster et al.
(2010) and co-workers.
5. Concluding remarks
Through incorporating new kinematical DOFs, the present work
has proposed an elastodynamic homogenization theory generalizing the one of Willis in the case of periodic media and reducing the error committed during the upscaling process, especially
at high frequencies. In order to illustrate the potential of the presented theory, it has been shown that the LW-LF asymptotic expansion of the effective motion equation is capable of simultaneously capturing the acoustic and the rst optical branch of the
microscopic dispersion curve for a simple 1D medium.
Two problems remain open. The rst concerns the effective
elastodynamic constitutive law produced by the generalized theory
proposed. In this paper, to avoid the diculty related to its nonuniqueness, the effective motion equation has been directly treated
and exploited. However, in numerous situations, it is useful and
important to explicitly know the effective elastodynamic constitutive law. The second problem regards the optimal choice of shape
functions for which guiding criteria exist but remain incomplete.
References
Andrianov, I.V., Bolshakov, V.I., Danishevskyy, V.V., Weichert, D., 2008. Higher order
asymptotic homogenization and wave propagation in periodic composite materials. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences 464, 11811201.
Antonakakis, T., Craster, R.V., Guenneau, S., 2014. Homogenisation for elastic photonic crystals and dynamic anisotropy. J. Mechanics Phys. Solids 71, 8496.
Auriault, J.-L., Bonnet, G., 1985. Dynamics of periodic elastic composites (Dynamique
des composites elastiques periodiques). Archiwum Mechaniki Stosowanej 37 (4),
269284.
Auriault, J.-L., Boutin, C., 2012. Long wavelength inner-resonance cut-off frequencies
in elastic composite materials. Int. J. Solids Struct. 49 (23-24), 32693281.
Bensoussan, A., Lions, J.L., Papanicolaou, G., 1978. Asymptotic analysis for periodic
structures. North-Holland Publishing Company.
Boutin, C., Auriault, J.-L., 1993. Rayleigh scattering in elastic composite materials.
Int. J. Eng. Sci. 31 (12), 16691689.

Boutin, C., Rallu, A., Hans, S., 2014. Large scale modulation of high frequency waves
in periodic elastic composites. J. Mech. Phys. Solids 70, 362381.
Colquitt, D.J., Craster, R.V., Antonakakis, T., Guenneau, S., 2014. Rayleigh-Bloch waves
along elastic diffraction gratings. Proc. Royal Soc. A: Math. Phys. Eng. Sci. 471,
20140465.
Craster, R.V., Kaplunov, J., Pichugin, A.V., 2010. High-frequency homogenization for
periodic media. Proc. Royal Soc. A: Math. Phys. Eng. Sci. 466, 23412362.
Daya, E., Braikat, B., Damil, N., Potier-Ferry, M., 2002. Continuum modeling for the
modulated vibration modes of large repetitive structures. Comptes Rendus Mcanique 330, 333338.
Fietz, C., Shvets, G., 2010. Current-driven metamaterial homogenization. Physica B:
Condensed Matt. 405 (14), 29302934.
Forest, S., 2006. Milieux continus gnraliss et matriaux htrognes. Collection
Sciences de La Matire. Presses de lEcole des Mines.
Forest, S., Sab, K., 1998. Cosserat overall modeling of heterogeneous materials. Mechanics Res. Commun. 25 (4), 449454.
Gazalet, J., Dupont, S., Kastelik, J.C., Rolland, Q., Djafari-Rouhani, B., 2013. A tutorial survey on waves propagating in periodic media: electronic, photonic and
phononic crystals. Perception of the Bloch theorem in both real and Fourier domains. Wave Motion 50 (3), 619654.
Germain, P., 1973. The method of virtual power in continuum mechanics. Part 2:
Microstructure. SIAM J. Appl. Math. 25 (3), 556575.
Nassar, H., He, Q.-C., Auffray, N., 2015a. On asymptotic elastodynamic homogenization approaches for periodic media. J. Mech. Phys. Solids (Submitted).
Nassar, H., He, Q.-C., Auffray, N., 2015b. Willis elastodynamic homogenization theory
revisited for periodic media. J. Mech. Phys. Solids 77, 158178.
Nolde, E., Craster, R.V., Kaplunov, J., 2011. High frequency homogenization for structural mechanics. J. Mech. Phys. Solids 59 (3), 651671.
Qur, Y., 1988. Physique des matriaux: cours et problmes. X Ecole Polytechnique.
Edition Marketing.
Sanchez-Palencia, E., 1980. Non-homogeneous media and vibration theory. SpringerVerlag.
Smyshlyaev, V.P., 2009. Propagation and localization of elastic waves in highly
anisotropic periodic composites via two-scale homogenization. Mech. Mater. 41
(4), 434447.
Smyshlyaev, V.P., Cherednichenko, K.D., 20 0 0. On rigorous derivation of strain gradient effects in the overall behaviour of periodic heterogeneous media. J. Mechanics Phys. Solids 48, 13251357.
Srivastava, A., Nemat-Nasser, S., 2014. On the limit and applicability of dynamic homogenization. Wave Motion 51 (7), 10451054.
Willis, J.R., 1997. Dynamics of composites. In: Suquet, P. (Ed.), Continuum Micromechanics. Springer-Verlag New York, Inc., pp. 265290.
Willis, J.R., 2011. Effective constitutive relations for waves in composites and metamaterials. Proc. Royal Soc. A: Math. Phys. Eng. Sci. 467 (2131), 18651879.
Willis, J.R., 2012. The construction of effective relations for waves in a composite.
Comptes Rendus Mcanique 340 (4-5), 181192.